Can we beat wrk?

code review
1

Just for fun, holiday project. I was trying to beat wrk for simple get request.
Disclaimer: i was using claude code for prototyping and optimization.

Results are here: rwrk/PERFORMANCE.md at master · arch7tect/rwrk · GitHub

key findings:

  • reqwest is slower then row hyper ~4-5%
  • wrk is very fast, but it depends heavily on the number of threads and connections
  • the number of Rust workers has a significant impact on the result

config:
Platform: macOS, 16 cores

rwrk: 56x CPU cores = 896 async workers (default)
./target/release/rwrk -u https://exposeme.org/ -t 30

wrk: 64 threads, 704 connections (optimal)
wrk -t64 -c704 -d30s https://exposeme.org/

current results:
rwrk - 7,145 req/sec
wrk - 7,143 req/sec

PS: I couldn't solve a problem of correct MB/sec counting. wrt reads from a raw socket so it is easy to measure total bytes. But I couldn't find a way to count MB/sec (including headers) in the hyper client.
PPS: updated figures with benchmark series in controlled environment.

2

wrk is a modern HTTP benchmarking tool capable of generating significant load when run on a single multi-core CPU. It combines a multithreaded design with scalable event notification systems such as epoll and kqueue.

3

absolutely correct. my question was, can async rust beat wrk in performance?

4

This topic was automatically closed 90 days after the last reply. We invite you to open a new topic if you have further questions or comments.